Release Notes¶

0.1.0 (2026-03-17)¶

Initial release.

Prompt scanning via beforeSubmitPrompt Cursor hook
Response scanning via afterAgentResponse Cursor hook with code extraction
Three modes: observe, enforce, bypass
Six detection services: prompt injection, DLP, toxicity, malicious code, URL categorization, custom topics
Per-service enforcement: block, mask, or allow independently
Fail-open design: never blocks on infrastructure failures
Circuit breaker: automatic bypass after consecutive API failures
DLP masking: replace sensitive content instead of blocking
Code extraction: fenced, indented, and heuristic detection
Structured logging: JSON Lines with automatic rotation at 10MB
Stats CLI: scan totals, block rates, latency percentiles
Global hook installation: --global flag for all Cursor workspaces
Precompiled JS: ~800ms cold start vs ~2.5s with tsx
Environment variable defaults: only AIRS_API_KEY required
66 tests across 9 suites including compiled JS integration tests